With the increasing integration of chatbots into educational settings, their role in supporting children’s collaborative creativity remains underexplored. This study investigates how children aged 9–12 interact with a chatbot during a real-world maker task and what challenges emerge in the process. Using a qualitative approach, we conducted four workshop-based group activities in a primary school setting, where children used the Grove Zero platform and the Doubao chatbot to collaboratively design prototypes for real-life problems. Data were collected through field observations and semi-structured interviews and analyzed using thematic analysis. The findings reveal three key challenges: (1) breakdowns in communication caused by children’s vague expressions; (2) disrupted collaboration due to changes in group roles and reduced peer interaction; and (3) growing behavioral dependence on chatbot over time. Based on these insights, we propose three design strategies: guided language scaffolding, collaboration-aware interaction design, and heuristic-based chatbot responses. This study offers practical implications for the development of child-centered educational AI systems.